depthfirst

depthfirst is a San Francisco-based applied AI lab that builds AI-native software security technology. Its platform, marketed as General Security Intelligence, uses autonomous agents that build context on a customer's code, infrastructure and business logic in order to detect, triage and remediate vulnerabilities across software and infrastructure layers, and to deliver ready-to-merge fixes inside developer workflows. The company also states the platform covers credential exposure and monitoring of threats in open source and third-party components.

The company's stated thesis is that AI-assisted coding has increased the volume and complexity of code being shipped while attackers have become automated and always-on, leaving legacy scanning and static-analysis tooling unable to keep pace. Rather than adapting general-purpose LLMs, depthfirst builds domain-specific security models, and it argues that owning the post-training process yields better task performance and lower inference cost. In March 2026 it introduced dfs-mini1, its first in-house security model, built on an open source base model and post-trained with reinforcement learning in security-specific environments, initially targeted at vulnerabilities in cryptocurrency smart contracts and evaluated on the OpenAI EVMBench benchmark. The company says internal evaluations indicate the training approach generalizes to other security domains, and that further vertical-specific models are planned.

depthfirst was founded in 2024 by Qasim Mithani (co-founder and CEO), Andrea Michi (co-founder and CTO) and Daniele Perito. Proceeds from its 2026 rounds are earmarked for training additional security models, expanding the applied AI research team, hiring across engineering, product and sales, and scaling enterprise adoption.

Founding story

depthfirst was founded in 2024 (TechCrunch reports October 2024) by three technical leaders from Google DeepMind, Databricks and Faire. Andrea Michi spent nearly seven years at Google DeepMind, where he was a Research Engineering Lead working on reinforcement learning for Gemini, AlphaDev and energy-efficiency projects for windfarms and data centers, and concluded that the most effective way to build AI for large-scale problems is to build purpose-specific models from the ground up rather than adapt general LLMs; he left DeepMind in 2024 to apply that approach to cybersecurity. Co-founder and CEO Qasim Mithani built developer and security products in the early days of AWS and was an early leader at Databricks, where he built infrastructure and the enterprise platform and served as head of infrastructure security. Co-founder Daniele Perito co-founded wholesale marketplace Faire, was a founding team member at Cash App leading security and fraud prevention (also described as director of security and risk engineering at Square/Block), and was previously a postdoctoral security researcher at UC Berkeley. The founders' shared premise was that they had seen how vulnerabilities emerge and spread and why existing approaches fall short.

Business model

depthfirst sells its AI-native security platform to businesses, including Fortune 500 enterprises and fast-growing software companies, positioning it as a replacement for legacy security scanning programs. It combines a customer-facing product with in-house specialized security models, aiming to own both the product experience and the underlying model intelligence.

Traction

Products became generally available in late 2025. As of January 2026, depthfirst reported agents uncovering 8x more true-positive vulnerabilities than traditional static analysis tools with an 85% reduction in false positives, and state-of-the-art results on CyberGym. As of March 2026, it reported more than 20 customers, revenue growth of 300% over the prior two quarters, and that 80% of its fix recommendations are accepted and merged by developers. Named customers include ClickUp, Lovable, Supabase, incident.io, Moveworks and AngelList.

Latest developments

On March 31, 2026, depthfirst announced an $80M Series B led by Meritech Capital at a $580M valuation, taking total funding to $120M, and launched dfs-mini1, its first in-house security model, initially focused on cryptocurrency smart contracts and evaluated on OpenAI EVMBench. The company said it will use the capital to train additional security models in new domains, expand its AI research team and scale enterprise adoption, and that further vertical-specific models will follow.